Credit rating migration risk and interconnectedness...
Transcript of Credit rating migration risk and interconnectedness...
Credit rating migration risk and interconnectedness in a
corporate lending networkI
Masayasu Kannoa,∗
aCollege of Commerce, Nihon University, Tokyo 157-8570, Japan
Abstract
This study assesses the credit rating migration risk and interconnectednessin Japan’s corporate lending market during the fiscal years 2008–2015. First,the study conducts a portfolio credit risk analysis by using outstanding lend-ing data with borrowers and lenders names. The results show an expectedshortfall with tail dependence of t-copula captures the heavy-tailed risk ofJapanese institutions. The study also measures credit risk exposures andcredit risk amounts by industry sector, and evaluates sector concentrationrisk. Subsequently, the study analyzes the network structure of lending con-tracts using network centrality measures. From the perspective of network,institutions play a central role in terms of degree centrality. Further, thestudy undertakes a stress test using a historical economic scenario pertain-ing to a credit rating migration matrix shortly after the Lehman Brothers’bankruptcy. The test finds a much sparser network structure because of alarge number of firm defaults. The study’s analysis offers banks and insurersimportant implications regarding the credit risk management of corporatelending.
Keywords: credit rating migration risk; sector concentration;interconnectedness; centrality measure; credit value at risk and expectedshortfall; stress testJEL classification: G32; G10; D85; L14; G28; F37.
IThis study was supported by the Japan Society for the Promotion of Science (JSPS)KAKENHI [Grants-in-Aid for Scientific Research, 17K03813]. This assistance is sincerelyappreciated.
∗Corresponding author. E-mail address: [email protected]
1. Introduction
After the collapse of the bubble economy in Japan, Japanese banks suf-fered because of efforts to dispose of non-performing loans. In this regard,the “lost decade” was a term initially coined to describe the Japanese econ-omy in the last decade of the prior millennium. However, the bursting of thehuge real estate bubble in Japan in the 1980s led to sluggish performancenot just in the subsequent “lost decade” but also in the first decade of thenew millennium. The term “lost decade” has also been used to describe thestate of the US economy from 2000 to 2009 because an economic boom inthe middle of this period was not enough to offset the effects of two hugerecessions.
In the past, banks in Japan have traditionally focused on the disposal ofnon-performing loans and compliance. Hence, the supervisory authority, theFinancial Services Agency of Japan, conducted rigorous asset assessmentsbased on certain criteria and inspected banks for compliance violations. Bycontrast, from the perspective of insurers, lending was undertaken as partof portfolio investments as well as shareholdings. However, because insurershold large numbers of shares, the proportion of lending in their portfolios isrelatively small.
Currently, internationally active banks in Japan are regulated based onthe Basel III framework (BCBS, 2005). Most major banks, including megabanks and other large banks, adopt an internal ratings-based (IRB) approach.This approach calculates risk-weighted assets in terms of firms’ lending assetsin accordance with the firms own obligor credit risks. By contrast, insuranceregulation in Japan depends on the Japanese local supervisory framework,which is based on the “solvency margin ratio.” This framework is simple;moreover, it is a so-called first-generation solvency regulation, which is sim-ilar to Basel I in international banking regulations. The “solvency marginstandard” was introduced for both life and general insurance firms in fiscalyear (FY)1 1996. The solvency margin standard is calculated as the solvencymargin divided by half of the risk amount, expressed as a percentage.
Triggered by the bankruptcy of Lehman Brothers, many Western banksand insurers suffered significant capital losses, with some also recording im-pairment losses. By contrast, Japanese banks were hardly affected owingto their experience of non-performing loan disposal in the country’s bubble
1Japan’s fiscal year runs from April 1 to March 31.
2
economy. However, a network analysis of contractual relationships in thecorporate lending network has yet to be conducted, even though large firmsordinarily have accounts with 10 or so banks. Further, the bankruptcy, orcredit rating downgrade, of a firm causes an increase in credit risk; thus, it isimportant to analyze the credit rating migration risk in the lending network.
The current study contributes to the literature by providing quantitativeinsights into the corporate lending network, taking into account credit ratingmigration risk. The study assesses the interconnectedness in the networkthat comprises various types of banks, insurers, and firms by using credit riskexposure implied by the credit ratings during and after the global financialcrisis.
First, this study conducts a credit risk analysis using the credit ratingmigration approach. In this approach, credit risk exposure by bank or insureris examined, depending on credit migration. In addition, this study ranksthe industry sectors in terms of sector concentration risk measured by creditrisk exposure and credit risk amounts measured by VaR and ES.
Second, this study analyzes interconnectedness in the lending network us-ing various network centrality measures (Jackson, 2010; Kanno, 2015a). Thisnetwork analysis is based on the outstanding lending dataset for all listedJapanese firms by financial institution (i.e., bank or insurer) in the NikkeiNEEDS FinancialQUEST database provided by Nikkei Inc., a Japanese news-paper firm. The dataset covers almost all lending contracts among bank-to-listed firms and insurer-to-listed firms. Thus, it is to be noted that thedataset with borrowers and lenders names is quite rare and valuable in theworld.
Third, this study conducts a stress test to verify the resilience of theJapanese corporate lending market and the change of network structure. Inthis test, copula dependence is incorporated to investigate the tail depen-dence of portfolio value distribution.
In the rest of this paper, section 2 reviews the literature on lending andinterconnectedness in various financial networks. Section 3 contains a creditrisk analysis using some risk measures, while section 4 presents a networkstructure analysis of lending market. Section 5 conducts a stress test bycombining both analyses and section 6 concludes the study.
3
2. Literature review
The current study contributes to the corporate lending literature by us-ing a combined credit risk and network analytical approach to investigatethe network structure of corporate lending contracts. Network analysis is ahighly effective approach to examine the interconnectedness of relationshipsin lending contracts. Such contracts represent complex contractual networksusing sets of “nodes” connected by “edges.” In a corporate lending network,a node represents a lender or a borrower, and an edge represents the lendingrelationship between two entities.
A large body of financial literature exists on corporate lending in countriesworldwide. However, the literature barely mentions the interconnectednessof corporate lending contracts in a country. Nonetheless, there are somestudies in this area such as Abbassi et al. (2017), De Masi and Gallegati(2012), Ha laj et al. (2015), Lux (2016), and Silva et al. (2018).
Abbassi et al. (2017) analyze the relationship between market-basedcredit risk interconnectedness among banks during the global financial crisisand the associated balance sheet linkages via funding and securities hold-ings. In this regard, the authors use a German data set that has the in-terbank funding positions for 2006–2013, together with the investments ofbanks at the security level and the credit register. De Masi and Gallegati(2012) use a database of Italian firms provided by Bureau van Dijk to un-dertake an empirical assessment of the credit relationships between banksand firms. However, because contractual amounts are not included in thedatabase, credit risk exposure is not captured; thus, a credit risk analysis isnot conducted.
Ha laj et al. (2015) use network formation techniques based on a theoreti-cal framework to construct networks of lending relationships between a largesample of banks and nonbanks in the European Union (EU). Lux (2016)employs a stochastic network model to review basic stylized facts found inthe comprehensive data sets of bank–firm loans for a number of countriesin order to consider credit linkages between banks and non-financial firms.Silva et al. (2018) simulate shocks to the real sector and evaluate how thefinancial system reacts; they then amplify these events using loan-level datain the Brazilian bank-bank (interbank) network and the bank-firm lendingnetwork.
In addition, although the structure of syndicated lending differs fromcorporate lending in terms of contractual features and data availability for
4
researchers, there is literature on the interconnectedness of the syndicatedlending market, such as Goldlewski et al. (2012), Wang and Wang (2012),and Wu et al. (2013). Goldlewski et al. (2012) use a data set of the Frenchsyndicated lending market from the DealScan database that includes de-tailed information on loan agreement and bank syndicate characteristics.This database is commonly used in empirical studies on syndicated lending.Wu et al. (2013) conduct theoretical and empirical investigations of the in-teractions among potential lenders and how these may influence contractualterms via informational cascade in the syndicated loan market.
Further, in terms of the use of network measures in an interfirm net-work, although somewhat different from a lending network, Garmaise andMoskowitz (2003) find they impact the availability of credit, which is vitalfor firms engaged in innovative activities. They affirm that firms need to havean appropriate level of financial incentives to encourage investments in long-development risky R&D intensive projects that risk-averse managers mightnot be willing to undertake. Chuluun et al. (2017) examine how variousdimensions of an interfirm network affect innovation and pricing of innova-tion by market participants. They construct a set of network measures thatcapture a firm’s centrality in an interfirm network (degree, eigenvector, andbetweenness), cohesion and diversity within firm networks (density, networknon-redundancy, and industry diversity), and innovativeness and propinquityof firm networks (network innovativeness and industry, geographic, innova-tive industry, and innovative geographic propinquity). Using these networkmeasures, they assess if firm network characteristics impact innovation inputand output.
An analytically tractable example of financial networks is the interbanknetwork characterized by bilateral exposure in the interbank market. Inthis context, studies of financial networks adopt two approaches. The firstassesses the strength of contagion channels and network resilience by observ-ing the responses of financial network structures to shocks. Introducing ashock assumes a specific transmission mechanism, such as defaults by coun-terparties. Alves et al. (2013) refer to this approach as “dynamic networkanalysis.” Cocco et al. (2009), Elsinger et al. (2006), and Haldane and May(2011) analyze contagion effects in their network analyses.
The second approach describes network structures using topological indi-cators, often relating these to model graphs based on network theory. Thisapproach does not assume a mechanism by which shocks propagate withinthe network; thus, it is referred to as “static network analysis” (Alves et al.,
5
2013). The studies of Boss et al. (2004), Eisenberg and Noe (2001), andKanno (2015a, 2015c, 2018a) are examples of this approach. The currentstudy adopts static network analysis. While many different centrality mea-sures exist, most of them apply to static networks. The details of centralitymeasures are described later in Section 4.2.
3. Credit risk analysis
This study analyzes credit risk based on the credit rating migration ap-proach, using a large-scale Japanese lending database.
3.1. Credit risk exposure for lending contracts
3.1.1. Methodology for credit risk exposure analysis
Credit rating migration is an essential component in credit portfolio val-uation. This study outlines a framework for gauging the effects of creditrating migration on portfolio risk measurements. The approach is based ondiscounted cash flow valuation, whereby a lending asset is valued by discount-ing the expected cash flow at a discount rate adjusted for credit risk. Therisk adjustment here can take the form of a higher discount rate. Discountrates adjusted for credit risk are obtained from credit rating curves providedby credit rating agencies such as Moody’s and Standard & Poors.
Throughout this study, the filtered probability space, (Ω,F ,Ft, Q), isalso incorporated, thereby supporting the credit rating migration processin terms of discrete time, t = 0, 1, . . . , T , where Q is a physical probabilitymeasure and the horizon, T , is assumed to be a positive integer indicating thematurity. The filtration, Ft, models the flow of all the observations availableto lenders. Formally, given an initial rating, C0, of a borrower, future changesin the rating are described by a stochastic migration process, C.
This study assumes that the set of rating classes is 1, . . . , K, wherethe state, K, is assumed to correspond to the default event. In addition,according to the convention of Jarrow et al. (1997), the order of the statesis fixed so that the state, j = 1, represents the highest ranking, whereas thestate, j = K − 1, represents the lowest non-default ranking.
With regard to lending exposures that are not in default, the theoreticalprice of a lending asset with certain future cash flow at time t is expressed
6
as an aggregate discounted present value, P , as follows:
P = EQ
[T∑t=1
CFt
(1 + r(Cit))
t
∣∣∣∣∣Ft
], (1)
where E means taking an expectation under a physical probability measure,Q, and the lending type corresponds to a term loan of equal monthly pay-ments with interest. Thus, maturity, T , corresponds to three years in thecase of city banks and trust banks and five years in other cases. CFt is cashflow scheduled at time t. r(Ci
t) is a discount rate adjusted for credit risk withregard to the rating Ci
t at time t provided by a rating agency i.
3.1.2. Data for credit risk exposure analysis
For the purpose of credit risk analysis, this study calculates the creditrisk exposure of a lending contract, discounting its cash flow at a discountrate adjusted for credit risk. To this end, the study uses firm-level outstand-ing lending contracts and financial data for FY2008–FY2015. The analysisrequires outstanding data with borrowers and lenders names. These are ob-tained from the Nikkei NEEDS FinancialQUEST database (Table 1). Thedatabase contains lending information on bank-to-listed firms and insurer-to-listed firms. Large but non-listed firms are not included.2 Thus, the coverageratio of large firms in the database may not be that high overall. The banksinclude city banks, trust banks, Shinsei Bank and Aozora Bank, NorinchukinBank, regional banks (i.e., regional banks I), second-tier regional banks (i.e.,regional banks II), Shinkin banks and credit unions, other private financialinstitutions, government financial institutions, and foreign banks. The in-surers include life insurers and non-life insurers. Finally, a certain amountof data for the lending contracts of unknown institutions is included in thedatabase.
In addition, this study uses average interest rates for new lending con-tracts by bank type (i.e., city banks, regional banks I/II, and Shinkin banks)from the Bank of Japan. As shown in Figure 1, city banks set interest ratesthat are higher in the long term than the short term, whereas the other banksadopt a reverse approach. Further, after the global financial crisis, long- and
2Known examples of non-listed large firms are Suntory Holdings (a beverage productsfirm), the Takenaka Corporation (a general construction firm), and Yanmar (an agricul-tural machinery manufacturing firm).
7
short-term interest rate levels decreased year by year. Interest rate levelsalso fell after the Bank of Japan initiated a “negative interest rate policy”in February 16, 2016. As a result, in terms of the outstanding lending ofdomestic banks in Japan, the lending share with interest rates less than 1%reached 62% of the entire outstanding lending, according to the financialnewspaper Nihon Keizai Shimbun, on February 16, 2018.
In terms of credit rating information, this study also uses credit ratinghistorical data, including “date of change” and “old and new credit ratings”by entity from the Nikkei Astra Manager database provided by the QUICKCorporation. The data concern long-term issuer ratings related to the cer-tainty of fulfillment of issuers’ individual financial obligations, as promised.However, not all listed firms are endowed with a credit rating. Thus, for suchfirms, outstanding lending is substituted for credit risk exposure.
Further, this study employs yield curves by credit rating obtained as a“credit rating matrix” from the homepage of the Japan Securities Dealers As-sociation (JSDA).3 Yield by credit rating means the mathematical average ofthe compound interest yield for over-the-counter (OTC) bond transactions,calculated using the quotations reported to the JSDA. As shown in Figure2, yield curves are provided for each business day by four credit rating agen-cies: Rating and Investment Information (R&I), Japan Credit Rating Agency(JCR), Moody’s, and Standard & Poors (S&P).
However, Japanese financial institutions and scholars highlight two ratedifference issues among the four credit rating agencies. One issue is thedifference in the approaches of the Japanese credit rating agencies (R&I andJCR) and the American credit rating agencies (Moody’s and S&P), which isthe equivalent of “two notches.” The other is the difference in the approachesof the R&I and JCR, which is the equivalent of “one notch.” In order tocorrect these differences, this study adopts the lowest credit rating when twoor more different credit ratings are assigned to a firm.
3.1.3. Results of credit risk exposure analysis
The estimation results of credit risk for lending contracts are now dis-cussed. Table 2 reports the quartiles, and mean and standard deviations, in
3This body is an association that functions as a self-regulatory organization and in-terlocutor between market participants and various stakeholders, including governmentauthorities. JSDA members consist of securities firms and other financial institutionsoperating securities businesses in Japan.
8
Figure 1: Average lending interest rate curves: Long-term lending (mixedcolor) and short-term lending (red) for the end of March 2009 to March 2017
Notes: The two graphs show short- and long-term interest rate yield curves by banktype for the end of January 2009 to August 2017. The indices are as follows. 1:domestic banks in Japan, including city banks, Shinsei Bank, Aozora Bank, trustbanks, regional banks I/II, and Shinkin banks; 2: city banks; 3: regional banks I;4: regional banks II and Shinkin banks.
the upper tier, and the outstanding sums by entity in the lower tier, thatare related to credit risk exposure at the end of the period FY2008–FY2015.In addition, Figure 3 illustrates the percentile distribution of bilateral creditrisk exposure by year.
As can be seen in the upper tier of Table 2 and in Figure 3, all of theexposure sizes are small at the median (i.e., the 50th percentile); however,the sizes increase sharply from the 99.5th percentile to the maximum, andrange from 2 119 billion Japanese yen (JPY) in FY2008 to a maximum ofJPY 3 881 billion in FY2010. This finding means that, for the purpose ofreducing credit risk exposure, corporate lending decreased sharply just afterthe Lehman Brothers’ bankruptcy. By contrast, since FY2011, outstandinglending has increased by 3% to 50% from FY2010.
Further, as can be seen in the lower tier of Table 2, the credit risk exposuresize in the entire network remain almost unchanged from FY2008 to FY2009
9
20082010
20122014
2016
0
10
200
1
2
3
fiscal year-end
R&I-AA
maturity
rate
20082010
20122014
2016
0
10
200
1
2
3
4
fiscal year-end
R&I-A
maturity
rate
20082010
20122014
2016
0
10
200
2
4
6
8
10
fiscal year-end
R&I-BBB
maturity
rate
20082010
20122014
2016
0
10
200
1
2
3
fiscal year-end
JCR-AAA
maturity
rate
20082010
20122014
2016
0
10
200
1
2
3
fiscal year-end
JCR-AA
maturity
rate
20082010
20122014
2016
0
10
200
2
4
6
fiscal year-end
JCR-A
maturity
rate
Figure 2: Credit yield curve
Notes: The six panels show credit rating curves for the end of March 2009 to March2016. The AA to BBB ratings of R&I extend from the upper-left panel to theupper-right panel. The AAA to A ratings of JCR extend from the lower-left panelto the lower-right panel.
despite the global financial crisis. In addition, major banks and other largebanks have a share of 50% to 59% of the total amount less the unknowns fromFY2008–FY2015, an effect that is particularly large in the Japanese lendingmarket. By contrast, regional banks (I and II) have an almost constant shareof 9% to 12%, and insurers have a share of 8% to 14% during this period.
3.2. Portfolio credit risk
This study conducts an analysis of lending portfolio credit risk by insti-tution based on value at risk (VaR) and expected shortfall (ES).
3.2.1. Methodology for portfolio credit risk analysis
This study considers a portfolio of lending exposures with a set of firmsas counterparties. It then conducts a copula-based, multifactor simulation of
10
Table 1: Lending exposures and other related variables
Item Description Sources
Lendings out-standing
Data on bilateral lending relations, suchas bank-to-listed firm and insurer-to-listed firm,
Nikkei NEEDSFinancial QUEST
Lending inter-est rates
Average interest rates for lendings out-standing by bank type, drawn in the Fig-ure 1
Bank of Japan
Credit ratings Credit rating history data including both‘date of the change’ and ‘old and newcredit ratings’ by entity
Nikkei AstraManager
Yield curves bycredit rating
Yield curves added credit risk premiumby rating assigned by four credit ratingagencies, partly as shown in the Figure 2
JSDA
Figure 3: Credit risk exposure distribution for corporate lending
Notes: Exposure amounts are expressed in JPY billions. The distribution shows therange from the 98th percentile to the 100th percentile.
11
Tab
le2:
Des
crip
tive
stat
isti
csof
outs
tan
din
gb
ilat
eral
cred
itri
skex
pos
ure
by
inst
itu
tion
(in
JP
Yb
illi
on)
2008
2009
2010
2011
2012
2013
2014
2015
25%
261
254
233
210
232
205
207
203
Med
ian
852
804
766
724
811
787
772
748
75%
2,85
52,67
72,54
32,44
63,00
23,01
62,92
22,95
4Max
imum
2,12
9,80
42,14
8,82
73,89
0,28
13,85
5,54
43,53
6,60
93,39
5,24
53,13
2,39
43,11
3,03
8Mean
6,92
96,81
97,12
67,25
18,44
710
,355
10,545
10,570
S.D
.45
,793
47,901
56,770
59,471
65,772
85,854
83,514
78,388
Citybanks
22,155
19,770
19,302
19,784
20,858
15,293
16,649
18,877
Trust
ban
ks
6,71
36,34
06,20
15,77
17,23
55,92
95,85
16,30
9Shinseiban
k&
Aozora
ban
k96
879
159
656
088
072
683
185
6Norinchukin
ban
k2,20
42,19
32,00
82,09
32,07
21,34
51,35
31,53
5Regionalbanks(I
&II)
6,31
25,80
05,46
25,46
25,10
84,00
24,86
14,82
8Shinkin
ban
ks&
Credit
unions
365
432
382
329
429
294
305
295
Other
private
FIs
4,31
14,04
43,98
44,34
34,99
53,66
03,62
43,79
3GovernmentFIs
5,09
05,84
27,45
56,08
66,26
85,23
84,81
24,39
5Other
foreignbanks
138
113
9266
3528
3435
Lifeinsurers
7,91
26,65
67,05
17,00
66,58
84,19
53,34
03,31
9Non
-lifeinsurers
8283
9066
8572
8098
Unknow
ns
37,752
39,841
41,810
46,306
43,019
65,172
93,804
92,994
Total
94,001
91,905
94,433
97,874
97,572
105,95
213
5,54
413
7,33
4
Notes:
S.D
.is
stan
darddeviation
andFIs
arefinan
cial
institutions.
Regionalbanks(I
andII)includeregionalbanksand
second-tierregion
alban
ks.
Threemegaban
ks,theMitsubishiUFJFinancialGroup,theMizuhoFinancialGroup,and
theSumitom
oMitsuiFinan
cial
Group,fallinto
thecategories
ofcity
banksandtrust
banks.
Acategory
of
“unknow
ns”
includes
finan
cial
institution
swithnam
esthatare
unknow
nontheNikkei
NEEDSFinancialQ
UEST
datab
ase.
12
credit rating migrations. Typical portfolio credit risk models are explainedin the literature, such as Crouhy et al. (2000), Gordy (2000), and Gupton etal. (1997).
Counterparty credit rating migrations and subsequent changes in port-folio values are calculated for each simulation scenario; moreover, some riskmeasurements are reported. Because a corporate lending contract has nomarket value, the portfolio value at the horizon is calculated by discountingfuture cash flow at a discount rate by credit rating, using equation (1). Inturn, the value for each counterparty’s lending exposure by scenario at therisk horizon is simulated based on the realized credit rating per counterparty.For example, in a given scenario, if a lending contract with a time to ma-turity of five years becomes a lending contract with a time to maturity offour years, the future cash flow of the contract is discounted by the discountrate of one-year forward four years maturity. Thus, the portfolio values ofshort-term exposures of less than one year are realized.
In order to incorporate the random variable by counterparty, this studyuses a multifactor model, associating each counterparty’s asset return with alatent random variable. This variable is mapped to a credit rating assignedfrom a credit quality at the horizon (Figure 4). Thresholds between creditratings at the horizon are calculated directly from a rating transition matrix.The model’s factors can depend on industry sectors such as construction andfood; geographical regions such as Japan, the United States of America, andthe eurozone; and any other credit risk driver. Each counterparty is assigneda series of weights that determine its sensitivity to each factor driving theunderlying credit risk.
This study defines M as the number of borrowers in a portfolio and K asthe number of systematic risk factors. Using a multifactor model, an assetreturn Ai (i = 1, . . . ,M) as a latent variable is then expressed as follows:
Ai =K∑k=1
wi,kZk +
√√√√1 −K∑k=1
w2i,kϵi, (2)
where Zk (k = 1, . . . , K) is a systematic risk factor associated with anunderlying credit driver, which is typically assigned for a specific indus-try or a domestic geographical region, and ϵi is firm i’s idiosyncratic riskfactor, which represents the firm-specific credit risk. The factor loadingwi,k (i = 1, . . . ,M ; k = 1, . . . , K) expresses a weight of an underlying sys-
13
tematic risk factor, k, for firm i. The total of the weights for each firm (i.e.,each row) is one. The weights are then calculated by regression analysis. Inaddition, this study assumes that a systematic risk factor, Zk (k = 1, . . . , K),and an idiosyncratic risk factor, ϵi, are all mutually independent.
By contrast, because a pair of systematic risk factors are mutually corre-lated, a correlation coefficient, ρi,j, between Ai and Aj, is expressed as
ρi,j = corr(Ai, Aj) =K∑k=1
K∑k′=1
wi,kwj,k′ ρk,k′ , (3)
where ρk,k′ is a correlation coefficient between Zk and Z ′k. This study then
assumes that a correlation coefficient between equity returns is a proxy vari-able of a correlation coefficient between asset returns. Further, a return fora TOPIX Sector Index to Zk is allocated; thus, a correlation matrix betweenthe systematic risk factors can be specified. If a correlation matrix betweenthe systematic risk factors cannot be specified, the factor correlation matrixdefaults to an identity matrix, meaning that the factors are not correlated.
With regard to each simulation scenario, the latent random variable, Ai,has a credit rating on the value distribution at the horizon (Figure 4). In turn,by using the credit rating curve, this study calculates the discounted valueat the horizon of future cash flow at a later date than that of the horizon.When the latent variables Ai (i = 1, . . . ,M) are normally distributed, thereis a Gaussian copula. An alternative structure is to let the latent variablesfollow a t distribution, which leads to a t copula. The degree of freedomfor a t copula controls the degree of tail dependence. The t copulas resultin heavier tails than Gaussian copulas. Implied credit correlations are alsolarger with t copulas. Switching between these two copula approaches canprovide important information on model risk (see, e.g., Cherubini et al.,2011).
Thus, we report risk measures such as VaR and ES for the value distri-bution at the horizon. First, in terms of VaR, given some confidence levelα ∈ (0, 1), the VaR of a portfolio at confidence level α is given by the smallestnumber x such that the probability that the loss X exceeds x is no largerthan (1 − α) as follows:
V aRα = infx ∈ R|P (X > x) ≤ 1 − α.
Second, for an integrable loss X and any α ∈ (0, 1), ES is the expected loss,
14
Figure 4: Portfolio value distribution at the horizon
given that the loss X is already beyond the pre-specified worst case levelV aRα as follows:
ESα = E(X|X ≥ V aRα).
3.2.2. Data for portfolio credit risk analysis
In order to calculate portfolio credit risk, the parameters inputted to themodel are shown in Table 3.
3.2.3. Results for portfolio credit risk analysis
Table 5 presents the results for credit risk amounts that are calculatedbased on VaR and ES for city banks, trust banks, Norinchukin Bank, majorregional banks, and major life insurers. VaR and ES incorporate dependentstructures among risk factors based on Gaussian copulas and t copulas withdegree of freedom five, respectively.
Consequently, by applying the same 99.9% confidence level to selectedinstitutions, Gaussian copula VaR, Gaussian copula ES, t copula VaR, and tcopula ES are ranked in ascending order for the institutions, except for MeijiYasuda Life and Dai-ichi Life. t copula ES is 3.48 times as large as Gaussiancopula VaR for the Mizuho Bank and 3.37 times larger than the average ofthe 16 institutions. As a result, ES requires more capital than VaR for the
15
Table 3: Inputs to portfolio credit risk model
Item Description Sources
Portfolio values Discounted present values for futurecashflows of lending contracts calculatedusing equation (1), if there is no data forcredit rating, lendings outstanding
JSDA
Ratings Credit rating migration by firm: pub-lished by four credit rating agencies
Nikkei AstraManager
Transition ma-trix
Matrix of credit rating transition proba-bilities with ratings as: ‘AAA’, ‘AA’, ‘A’,‘BBB’, ‘BB’, ‘B’, ‘CCC–C’, and ‘Default’as shown in the Table 4
R&I
LGD Loss given default for corporate lendingexposures: set to 45% evenly across allfirms.
Fundamentalinternal rating-based approachin Basel III
Weights Factor and idiosyncratic weights formodel
Confidence in-terval
Target for VaR and ES: set to 99.9% Own calculations
Factor correla-tion matrix
33 × 33 correlations among returns ofTOPIX Sector Indices
Nikkei AstraManager and owncalculations
Number of sce-narios
Set to 500,000 —
institutions. Unless the availability of risk measures is validated, there maybe inadequate credit risk management.
In addition, the three mega banks overwhelm other institutions in termsof credit risk amounts. This finding proves that the three mega banks havebeen selected as G-SIBs continuously since November 2011. For reference,core tier 1 capital by institution is shown in Table 5. In order to meet thecapital requirement in Basel III, internationally active banks raise additionalcapital by various instruments. For example, the Mitsubishi UFJ FinancialGroup issued senior bonds worth USD 5 billion in March 2016 and USD 2billion in April 2016 to meet total loss-absorbing capacity (TLAC) (BCBS,2016; Kanno, 2018a).
16
Table 4: Transition matrix with averaged R&I’s annual rating migrationrates for FY1978–FY2015
AAA AA A BBB BB B CCC-C DefaultAAA 91.00 9.00 0.00 0.00 0.00 0.00 0.00 0.00AA 0.70 94.40 4.80 0.10 0.00 0.00 0.00 0.00A 0.00 1.70 94.80 3.40 0.10 0.00 0.00 0.00BBB 0.00 0.00 3.80 93.50 2.60 0.00 0.00 0.10BB 0.00 0.00 0.20 8.10 86.50 2.60 0.10 2.50B 0.00 0.00 0.00 0.80 9.80 76.60 0.80 12.00CCC-C 0.00 0.00 0.00 0.00 0.00 6.50 87.00 6.50Default 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00
Figure 5 gives us additional information about the distribution of portfo-lio values for selected institutions and the selection of risk measures. Anorange vertical line corresponds to the current credit ranking. Thus, itshould be noted that the difference between the market and current valuecorresponds to the loss amount and that the loss distribution has the sameshape as the portfolio value distribution. The distributions of banks such asNorinchukin Bank and Yokohama Bank are asymmetrically unimodal distri-butions, whereas the distributions of institutions such as Chiba Bank andMeiji Yasuda Life are bimodal distributions and the distribution of Sumit-omo Life is trimodal distribution. In a multimodal distribution with two ormore modes, it is probable that any variation of VaR depends largely on theconfidence interval. By contrast, ES does not cause such an issue becausethe measure calculates the average of loss in the range beyond a confidenceinterval.
Because the average coverage pertaining to the outstanding lending datain the database is 27.29% for all city banks (five banks), 36.39% for majortrust banks (three banks), 5.35% for all regional banks I (64 banks), and17.84% (Cov1) for major life insurers (four banks), the risk amounts do notnecessarily represent each institution’s total lending credit risk.4 However,because the credit risk amounts for the three mega banks are JPY 159–223billion based on Gaussian copula VaR and JPY 657–828 billion based on t
4With regard to trust banks, outstanding lending by money trusts is not fully captured,whereas with regard to regional banks I, the percentage shares for non-listed large firmsare high.
17
copula ES, the figures need to be monitored carefully. The core tier 1 ratiofor each bank is at a high enough level for capital adequacy requirements.
3.3. Credit risk by industry sector
This section analyzes the credit rating migration risk in industry sectors,which are categorized into 33 sectors indexed by TOPIX Sector Indices. First,each firm’s credit risk exposure is aggregated in the industry sector to whichthe firm belongs. Table 6 indicates the ranking of the top 10 sectors inaccordance with credit risk exposure for FY2008–FY2015. Sectors such asWholesale Trade, Other Financing Business, Land & Transportation, ElectricPower & Gas, and Real Estate especially have large credit risk exposures.The information & Communication sector has been ranked in the top 10since FY2011, because during the period of FY2008–FY2015, the numberof firms has monotonically increased from 97 in FY2008 to 172 in FY2015and the total credit risk exposure for three major telecom carriers in Japan– Softbank Group, Nippon Telegraph & Telephone, and KDDI – has morethan doubled from JPY 1.8 trillion in FY2008 to JPY 4.1 trillion in FY2015.5
In addition, two sectors, Other Financial Business (mainly, lease firmsand credit card firms) and Real Estate, have higher credit risk owing to theirlarge financial assets. Since the end of 2015, lending to the real estate sectorhas started to increase. The real estate sector has especially increased itsoutstanding ranking because of the effect of the negative interest rate policyannounced by the Bank of Japan on January 29, 2016.
Further, since the Great East Japan Earthquake, which occurred onMarch 11, 2011 and the subsequent accident at the Tokyo Electric PowerCompany (TEPCO)’s Fukushima Daiichi Nuclear Power Plant, the TEPCOaccounts for a large proportion of credit risk exposure in the Electric Power& Gas sector for FY2010 (i.e., at the end of March 2011), with a suddenlarge increase in proportion from 25% for FY2009 to 39% in FY2010.
5In particular, to conduct strategic M&A or form an alliance, Softbank Group borroweda large amount of money, despite its S&P low credit rating of BB or BBB, incurring highinterest costs for the period. For example, SoftBank Group and Sprint Nextel (a NYSElisted firm) completed their merger. SoftBank Group invested approximately USD 21.6billion in Sprint, consisting of around USD 16.6 billion to be distributed to Sprint stock-holders and an aggregate of USD 5 billion of new capital to strengthen Sprint’s balancesheet. Sprint stockholders approved the transaction at a special meeting of stockholdersheld on June 25, 2013.
18
Tab
le5:
Cre
dit
risk
amou
nts
atth
een
dof
Mar
ch20
16(i
nJP
Ym
illi
on,
%)
Institutionname
IB/DBGC-V
aR
GC-E
St 5-C
-VaR
t 5-C
-ES
SME
lend-
ing
&Housing
loanratio
Large
firm
lending
ratio
Cov1
Cov2
CT1
MizuhoBank
IB223,169
335,775
534,140
828,288
59.49
40.51
12.50
30.86
11.20
BTMU
IB159,295
263,986
422,863
662,339
55.17
44.83
9.35
20.85
12.04
SMBC
IB182,301
266,236
438,885
657,177
67.90
32.10
10.88
33.90
13.15
ResonaBank
DB
32,088
48,418
86,631
124,705
82.32
17.68
5.72
32.36
10.58
SaitamaResonaBank
DB
6,083
9,441
9,663
13,322
87.83
12.17
0.87
7.17
11.58
Subtotal:
Citybanks
63.57
36.43
9.94
27.29
–MUTB
IB67,084
88,784
161,365
247,364
51.27
48.73
22.45
46.08
15.87
SMTB
IB98,154
145,625
250,653
374,527
57.68
42.32
13.48
31.86
10.76
MizuhoTB
IB96,717
143,002
224,847
346,465
50.00
50.00
13.77
27.54
18.73
Subtotal:
Tru
stbanks
55.82
44.18
16.08
36.39
–Norinch
ukin
Bank
IB41,131
55,364
107,991
164,692
––
10.72
–18.44
YokohamaBank
IB22,617
28,912
42,698
61,472
80.61
19.39
3.76
19.39
11.63
ShizuokaBank
IB8,321
11,403
13,890
20,178
77.68
22.32
1.45
6.49
16.35
ChibaBank
IB8,289
9,680
13,120
18,226
81.70
18.30
1.38
7.55
12.09
Subtotal:
Reg
ionalbanksI(64banks)
70.25
29.75
1.59
5.35
–NipponLife
–90,844
104,516
118,329
183,935
––
17.71
––
MeijiYasu
daLife
–58,065
100,986
76,366
138,383
––
18.07
––
Dai-ichiLife
–16,052
23,952
25,065
38,077
––
10.36
––
SumitomoLife
–16,719
20,252
23,131
33,718
––
11.38
––
Subtotal:
Majorlife
insu
rers
––
17.84
––
Note
s1:
IBis
anintern
ationallyactivebankin
accord
ance
withBaselIII;
DB
isabankth
atfocu
sesondomesticoperations;
GC
den
otes
Gaussianco
pula;t 5-C
den
otestco
pula
withdeg
reeoffreedom
five;
CT1is
core
tier
1ca
pital;Cov1is
theratioofoutstandinglendingin
thedatabase
tooneforea
chinstitution’s
financialstatemen
t;Cov2is
theratioofoutstandinglendingin
thedatabase
tooneforea
chinstitution’s
outstandinglendingto
largefirm
s.
Note
s2:
SME
&Housingloanratioreferto
themonth
lyKinyuJourn
alandco
retier
1refers
toth
estatisticsofth
eJapaneseBankers
Association.A
hyphen
indicatesnorelevantdata.Largefirm
sincludelisted
andnon-listedfirm
s.Theco
retier
1ratiois
notpublish
edfor
life
insu
rers.ResonaBankandSaitamaResonaBankhaveadoptedth
eca
pitaladeq
uacy
requirem
ents
forbanksth
atfocu
sondomestic
operations;
thus,
theirminim
um
capitalrequirem
entis
4%.
19
Fig
ure
5:D
istr
ibu
tion
ofp
ortf
olio
valu
esfo
rse
lect
edin
stit
uti
ons
(in
JP
Ym
illi
on)
Notes:
Thepan
elsshow
thedistribution
ofportfolio
values
byinstitutionattheendofMarch2016.Theupper
panel
by
institution
isfortheGau
ssiancopula
andthelower
panel
byinstitutionis
forthetcopula.Each
orangeverticalline
correspon
dsto
thevalueforthecurrentcredit
rating.Theinstitutionsare
MizuhoBank,theBankof
Tok
yo-M
itsubishiUFJ(B
TMU),Sumitom
oMitsuiBankingCorporation(SMBC),
andResonaBankfrom
the
upper-leftpan
elto
theupper-rightpan
el;Norinchukin
Bank,YokohamaBank,Shizuoka
Bank,andChibaBankfrom
themiddle-leftpan
elto
themiddle-rightpan
el;NipponLife,
MeijiYasudaLife,
Dai-ichiLife,
andSumitomoLifefrom
thelower-leftpan
elto
thelower-rightpan
el.
20
In turn, we calculate risk measures such as VaR and ES by sector forFY2015 (i.e., at the end of March 2016). Because the one-year forward port-folio value distribution based on the credit rating migration is derived byindustry sector, VaR and ES are calculated in relation to the distribution.Table 7 shows the ranking of the top 10 sectors in terms of credit risk as eval-uated by four risk measures. In general, the top 10 sectors are all similar tothe credit risk exposure for FY2015. However, the ranking for t5-Copula-ESis a little bit different from the others, because some of them have multimodaldistributions (Figure 6). Compared to the ranking for FY2015 in Table 6,sectors such as Electric Power & Gas, Chemicals, and Information & Com-munication are outside the ranking for all risk measures in Table 7. Althoughthese sectors have large credit risk exposure, the volatilities of portfolio valuedistributions are all small, as shown in the three lower panels of the Figure6. Hence, credit risk amounts measured by risk measures such as VaR andES are also small.
4. Network analysis
This section describes the study’s analysis of the network structures ofthe Japanese lending market in terms of bank-to-listed firms’ and insurer-to-listed firms’ relationships. The analysis is based on credit risk exposure.Such an approach differs from the nominal exposure that is examined in mostliterature on credit risk management.
4.1. Data for network analysis
The following (N ×N) matrix, X, represents Japanese corporate lendingrelationships:
X =
x11 · · · x1j · · · x1N...
. . ....
......
xi1 · · · xij · · · xiN...
. . ....
......
xN1 · · · xNj · · · xNN
, (4)
where xij denotes the outstanding exposure pertaining to firm i in terms ofthe lending of institution j. The summation across row i provides firm i’stotal outstanding exposure of its borrowing liabilities. The summation of
21
Tab
le6:
Top
10se
ctor
sra
nke
dby
cred
itri
skex
pos
ure
Rankin
g2008
2009
2010
2011
2012
2013
2014
2015
1W
holesa
leTra
de
Wholesa
leTra
de
Wholesa
leTra
de
Wholesa
leTra
de
Electric
Power
&Gas
Electric
Power
&Gas
Wholesa
leTra
de
Electric
Power
&Gas
2Oth
er
Fin
ancin
gBusiness
Oth
er
Fin
ancin
gBusiness
Oth
er
Fin
ancin
gBusiness
Electric
Power
&Gas
Wholesa
leTra
de
Wholesa
leTra
de
Electric
Power
&Gas
Wholesa
leTra
de
3Electric
Power
&Gas
Electric
Power
&Gas
Electric
Power
&Gas
Oth
er
Fin
ancin
gBusiness
Oth
er
Fin
ancin
gBusiness
Oth
er
Fin
ancin
gBusiness
Electric
Appli-
ances
Electric
Appli-
ances
4Land
Tra
nsp
orta-
tion
Land
Tra
nsp
orta-
tion
Land
Tra
nsp
orta-
tion
Land
Tra
nsp
orta-
tion
Land
Tra
nsp
orta-
tion
Land
Tra
nsp
orta-
tion
Oth
er
Fin
ancin
gBusiness
Oth
er
Fin
ancin
gBusiness
5RealEstate
RealEstate
RealEstate
RealEstate
RealEstate
RealEstate
Land
Tra
nsp
orta-
tion
RealEstate
6Electric
Appli-
ances
Tra
nsp
ortation
Equip
ment
Tra
nsp
ortation
Equip
ment
Electric
Appli-
ances
Chemicals
Inform
ation
&Communication
RealEstate
Land
Tra
nsp
orta-
tion
7Tra
nsp
ortation
Equip
ment
Electric
Appli-
ances
Electric
Appli-
ances
Tra
nsp
ortation
Equip
ment
Iron
&Ste
el
Chemicals
Tra
nsp
ortation
Equip
ment
Tra
nsp
ortation
Equip
ment
8Chemicals
Chemicals
Chemicals
Chemicals
Inform
ation
&Communication
Iron
&Ste
el
Inform
ation
&Communication
Chemicals
9M
achin
ery
Machin
ery
Machin
ery
Inform
ation
&Communication
Securities&
Com-
moditiesFutu
res
Securities&
Com-
moditiesFutu
res
Chemicals
Inform
ation
&Communication
10
Reta
ilTra
de
Reta
ilTra
de
Reta
ilTra
de
Machin
ery
Reta
ilTra
de
Reta
ilTra
de
Reta
ilTra
de
Reta
ilTra
de
Tab
le7:
Top
10se
ctor
sra
nke
dby
risk
mea
sure
sat
the
end
ofM
arch
2016
Rankin
gGaussian
Copula-V
aR
Gaussian
Copula-E
St 5
-Copula-V
aR
t 5-C
opula-E
S1
Electric
Appliances
Electric
Appliances
Electric
Appliances
Electric
Appliances
2Tra
nsp
ortation
Equip
ment
Tra
nsp
ortation
Equip
ment
RealEstate
Tra
nsp
ortation
Equip
ment
3RealEstate
RealEstate
Tra
nsp
ortation
Equip
ment
RealEstate
4Oth
erFin
ancin
gBusiness
Oth
erFin
ancin
gBusiness
Oth
erFin
ancin
gBusiness
Land,M
arine&
Air
Tra
nsp
ortation
5Land,M
arine&
Air
Tra
nsp
ortation
Land,M
arine&
Air
Tra
nsp
ortation
Land,M
arine&
Air
Tra
nsp
ortation
Oth
erFin
ancin
gBusiness
6W
holesa
leTra
de
Wholesa
leTra
de
Wholesa
leTra
de
Reta
ilTra
de
7Oil,
Coal,
Rubber,
Glass
&Cera
mics
Pro
ducts
Oil,
Coal,
Rubber,
Glass
&Cera
mics
Pro
ducts
Reta
ilTra
de
Wholesa
leTra
de
8Reta
ilTra
de
Reta
ilTra
de
Oil,
Coal,
Rubber,
Glass
&Cera
mics
Pro
ducts
Oil,
Coal,
Rubber,
Glass
&Cera
mics
Pro
ducts
9M
inin
gSecurities&
CommoditiesFutu
res
Foods
Foods
10
Foods
Pulp
&Paper
Securities&
CommoditiesFutu
res
Constru
ction
Note
s:Dueto
alimited
number
offirm
sin
somesectors,th
ose
sectors
are
merged
into
aggregate
sectors
asfollows:
Oil,Coal,Rubber,Glass
&CeramicsProductssector:
Oil&
CoalProducts,
Rubber
Products,
andGlass
&CeramicsProducts;
andLand,Marine&
Air
Transp
ortationsector:
LandTransp
ortation,MarineTransp
ortation,andAir
Transp
ortation.
22
Fig
ure
6:D
istr
ibu
tion
ofp
ortf
olio
valu
esfo
rse
lect
edse
ctor
s(i
nJP
Ym
illi
on)
Notes:
Thepan
elsshow
thedistribution
ofportfolio
values
bysectorattheendofMarch2016.Theupper
panel
bysector
isfortheGau
ssiancopula
andthelower
pan
elbysectoris
forthetcopula.Each
orangeverticallinecorrespondsto
thevalueforthecurrentcreditrating.
23
column j provides the total outstanding exposure of firm j’s lending assets.Thus, matrix X is asymmetric.
Because the analysis requires outstanding data for the credit risk exposurematrix, X, on lending relationships, this study utilizes the details by entity,as shown in Table 2.
4.2. Methodology and analytical results
This study calculates the network statistics and centrality measures forFY2008–FY2015 (see Table 8). Network size indicates the total number oflinks in the lending network. Table 8 shows that after FY2008, network sizeremains unchanged overall. This study also calculates four centrality mea-sures: degree centrality, eccentricity, hyperlink-induced topic search (HITS)hub centrality, and eigenvector centrality. Table 8 reports the averages foreach of these.
“Direct” centrality measures capture the level of interconnectedness ina local region, based on adjacent connections, and are proxies for lendinginfluence. These measures are degree centrality and eigenvector centrality.By contrast, “indirect” centrality measures enable the analysis of a counter-party’s exposure in the entire network in accordance with its distance to allother entities. These measures are used here to evaluate networks oriented toinformation value. Eccentricity is an example of indirect centrality measures.It shows how close an entity node is to other nodes in the entire network inorder to reflect the importance of one firm in the network (Renneboog andZhao, 2014).
It is important to understand that managerial influence and informationare two aspects of the same network. The two measures are not exclusive;the direct measures that express an entity’s managerial influence on its coun-terparties also have the ability to capture information, which could benefitthe entity. Nonetheless, the correlation between direct and indirect central-ity measures is generally low (Kanno, 2018b), suggesting that such measuresindeed capture different properties of the network. The two panels of Fig-ure 7 indicate “network size and direct centralities” and “exposure size andindirect centrality.”
4.2.1. Degree centrality
In terms of degree centrality, an entity’s total degree is the sum of itsin- and out-degrees. Because an institution is a lender, the institution hasonly one in-degree and no out-degree in terms of its relationship to a firm,
24
whereas a firm has only one out-degree and no in-degree in terms of itsrelationship to an institution. An entity’s degree is a proxy variable for itsinterconnectedness in the network. In a directed graph, all liabilities of aset of entities are directed from one borrowing firm to its lending institution.Degree centrality and network size are the same variables in a lending networkowing to the one-way transaction from an obligor to a creditor.
4.2.2. Eccentricity
Eccentricity is a measure of the maximum distance between a single entityand any other entity in the network. The distance, E(bi, bj), between theentities bi and bj is the sum of the edge weights expressed in the lendingcredit risk exposure on the shortest path from bi to bj in network G. Thus,the eccentricity of an entity bi is
E(bi) = arg maxbj∈G
d(bi, bj), (5)
where E(bi) ≥ 1.Table 8 shows that this centrality increases slowly for the period. In
addition, the correlation between eccentricity and HITS hub centrality is0.71 higher for the period, whereas the correlation between eccentricity anddegree centrality is −0.94 for the period (see Figure 7).
4.2.3. HITS hub centrality
In terms of HITS hub centrality, HITS is known as “hubs” and “authori-ties.” HITS was originally proposed to find the main structures in the WorldWide Web (WWW). Web pages are divided into two categories: hubs andauthorities. By the creation of a hyperlink from pages p to q, the authorof page p increases the authority of page q. The authority of a WWW sitewould consider its in-degree (i.e., the hyperlinks required to return to thehome page). Thus, HITS authority centrality is not suitable for measuringthe credit risk of a firm as a borrower in the lending network. By contrast, ahub is defined as a WWW site that indicates many authorities. Thus, HITShub centrality considers the credit risk of a borrower in terms of hub scoresbased on its out-degree. Institutions with the highest hub play a central rolein the network. The weights are normalized to ensure that the sum of theirsquares is 1.
25
4.2.4. Eigenvector centrality
Eigenvector centrality is a natural extension of simple degree centrality.Degree centrality awards one centrality point for every network neighbor of anentity. However, not all neighbors are equivalent. In many cases, an entity’simportance in a network increases owing to its connections to other importantentities. This defines the concept of eigenvector centrality (Newman, 2010).The advantage of eigenvector centrality over other centrality measures is thatit not only captures the number of entities linked to the target entity (degreecentrality); it also captures the centrality of the adjacent entities. Thus, anentity has a higher eigenvector centrality score if it is connected to moreentities with higher centrality scores.
Let Ce(g) denote the eigenvector centrality associated with network g. Anentity’s centrality is proportional to the sum of the centrality of its neigh-boring entities, λCe
i (g) =∑
j gijCej (g), for firm i. Using matrix notation,
λCe(g) = gCe(g), (6)
where λ is a proportionality factor. Thus, it can be seen that in Equation (6),Ce(g) is an eigenvector of g and λ is its corresponding eigenvalue. Becauseeigenvector centrality is a measure with nonnegative values, this study usesthe eigenvector associated with the largest eigenvalue (Jackson, 2010).
Table 8 shows that the average eigenvector centrality by year remains con-stant for FY2008–FY2012 and gradually decreases, together with the networksize, after FY2012.
Figure 7: Time-transitions of centralities
26
Table 8: Lending network structure based on credit risk exposure with bank–firm and insurer–firm relationships.
FY Network size Degree Eccentricity Hub Eigenvector2008 13,567 8.39 0.577 0.000309 0.003392009 13,461 8.32 0.583 0.000309 0.003392010 13,254 8.22 0.591 0.000310 0.003402011 13,475 8.38 0.602 0.000311 0.003452012 11,414 7.11 0.503 0.000311 0.003602013 9,755 6.08 0.524 0.000312 0.003812014 10,466 6.53 0.666 0.000312 0.002782015 10,475 6.53 0.681 0.000312 0.00270
Note: Network size is the total number of lending relationships in the network.
4.2.5. Ranking by degree
Table 9 shows the ranking of the top 20 entities in accordance with inter-connectedness, measured by the degree of their nodes. They include 15–17banks and 2–4 insurers. Because no firm borrows money from hundreds of in-stitutions, no firm ranks in the top 20. This table includes major banks suchas the Mitsubishi UFJ Financial Group (BTMU and MUTB), the MizuhoFinancial Group (Mizuho Bank, Mizuho Corporate Bank, Mizuho Trust &Banking), the Sumitomo Mitsui Financial Group (SMBC), Resona Holdings(Resona Bank), Sumitomo Mitsui Trust Holdings (SMTB) (Sumitomo Trust& Banking and Chuo Mitsui Trust & Banking prior to the merger); Nor-inchukin Bank; Government Financial Institutions (the Development Bankof Japan (DBJ) and Shoko Chukin Bank); major regional banks such asYokohama Bank, Fukuoka Bank, Chiba Bank, Joyo Bank, and Iyo Bank;and major life insurers such as Nippon Life, Meiji Yasuda Life, and Dai-IchiLife.
The degree centralities for financial institutions correspond to in-degreesin terms of the amount of borrowing by listed firms, whereas the degreecentralities for listed firms correspond to out-degrees in terms of the numberof lenders. However, in general, degree centralities, except for Mizuho Bank6
6The bank merged with Mizuho Corporate Bank on July 1, 2013.
27
and SMTB7 decrease gradually, as can be seen in Figure 8.Figures 9 and 10 offer a visual analysis by depicting directed graphs based
on degrees over 40 as at the end of March 2009 and the end of March 2016respectively. The direction of the arrow is from an obligor firm to a creditorinstitution. For example, BTMU has 1 236 in-degrees and 0 out-degrees inFY2008. As shown in Figures 9 and 10, because the edge is weighted byexposure, some thick ingoing edges flow into banks and life insurers fromfirms.
Figure 11 presents the six time-transition panels pertaining to a directedgraph based on degrees over 40 for FY2009–FY2014. The graphs show thatsome mega banks were exposed to large credit risk exposure originating fromthe Orix Corporation8 during FY2009–FY2012 and Kyushu Electric Power,an electric power firm, during FY2011–FY2014.
5. Stress test
This study conducts a stress test to verify the increase of credit risk interms of the deterioration of lending assets in the lending network at a riskhorizon in the future. Examples of the literature on stress tests of portfoliocredit risk are Breuer et al. (2012), Tsaig et al. (2011), and Varotto (2012).9
These studies’ tests differ from typical macro stress tests that consider theshocks of macroeconomic variables on risk parameters for each entity (Henryand Kok, 2013; Kanno, 2015a, 2015b).
By contrast, according to R&I (2016), the empirical probabilities of de-fault (PDs) pertaining to Japanese firms rated as speculative grades (i.e.,BB category or lower) by R&I reached a peak of 15% during the Heisei greatrecession (1997–1998) and during the two years following Lehman Brothers’bankruptcy (2008–2009). Consequently, this study’s test considers that thehistorical economic scenario pertaining to the credit rating migration matrixjust after Lehman Brothers’ bankruptcy is one of the worst scenario casesfaced by the Japanese economy, as shown in Table 10. Comparing Table
7Sumitomo Trust & Banking merged Chuo Mitsui Trust & Banking and Chuo MitsuiAsset Trust and Banking (non-listed) on April 1, 2012.
8This firm conducts a leasing business and has expanded into related fields such asbanking, insurance, and real estate.
9In addition, Schuermann (2014) lays out a framework for the stress testing of banksin terms of capital and liquidity.
28
Tab
le9:
Top
20en
titi
esra
nke
dby
inte
rcon
nec
ted
nes
s:D
egre
ece
ntr
alit
y
Rankin
g2008
2009
2010
2011
2012
2013
2014
2015
1BTM
U(1
236)
BTM
U(1
226)
BTM
U(1
208)
BTM
U(1
216)
BTM
U(9
87)
BTM
U(7
97)
UNKNOW
N(1
172)
UNKNOW
N(1
211)
2SM
BC
(1060)
SM
BC
(1042)
SM
BC
(1040)
SM
BC
(1063)
SM
BC
(876)
UNKNOW
N(7
91)
BTM
U(8
45)
BTM
U(8
06)
3UNKNOW
N(6
98)
UNKNOW
N(7
01)
UNKNOW
N(7
69)
UNKNOW
N(7
93)
UNKNOW
N(6
64)
MIZ
UHO
BANK
(770)
MIZ
UHO
BANK
(797)
MIZ
UHO
BANK
(801)
4M
IZUHO
BANK
(659)
MIZ
UHO
BANK
(667)
MIZ
UHO
BANK
(661)
MIZ
UHO
BANK
(667)
SUM
ITOM
OM
ITSUI
TRUST
BANK
(565)
SM
BC
(713)
SM
BC
(762)
SM
BC
(764)
5M
IZUHO
CORP
BANK
(535)
MIZ
UHO
CORP
BANK
(505)
RESONA
BANK
(503)
RESONA
BANK
(526)
MIZ
UHO
BANK
(541)
SUM
ITOM
OM
ITSUI
TRUST
BANK
(443)
SUM
ITOM
OM
ITSUI
TRUST
BANK
(449)
SUM
ITOM
OM
ITSUI
TRUST
BANK
(444)
6NIP
PON
LIF
E(5
11)
RESONA
BANK
(499)
MUTB
(491)
MUTB
(490)
RESONA
BANK
(457)
RESONA
BANK
(401)
RESONA
BANK
(417)
RESONA
BANK
(409)
7RESONA
BANK
(504)
NIP
PON
LIF
E(4
97)
NIP
PON
LIF
E(4
85)
MIZ
UHO
CORP
BANK
(477)
MUTB
(438)
MUTB
(347)
MUTB
(352)
NIP
PON
LIF
E(3
68)
8M
UTB
(499)
MUTB
(493)
MIZ
UHO
CORP
BANK
(483)
NIP
PON
LIF
E(4
59)
MIZ
UHO
CORP
BANK
(402)
NIP
PON
LIF
E(3
28)
NIP
PON
LIF
E(3
50)
MUTB
(335)
9SUM
ITOM
OM
ITSUI
TRUST
BANK
(404)
SUM
ITOM
OM
ITSUI
TRUST
BANK
(415)
SUM
ITOM
OM
ITSUI
TRUST
BANK
(415)
SUM
ITOM
OM
ITSUI
TRUST
BANK
(420)
NIP
PON
LIF
E(3
94)
DBJ
(255)
DBJ
(252)
DBJ
(248)
10
NORIN
CHUKIN
BANK
(375)
NORIN
CHUKIN
BANK
(350)
DBJ
(345)
DBJ
(357)
DBJ
(320)
NORIN
CHUKIN
BANK
(236)
NORIN
CHUKIN
BANK
(224)
NORIN
CHUKIN
BANK
(228)
11
CHUO
MIT
-SUI
TRUST
&BANKIN
G(3
41)
DBJ
(333)
NORIN
CHUKIN
BANK
(336)
NORIN
CHUKIN
BANK
(335)
NORIN
CHUKIN
BANK
(302)
SHOKO
CHUKIN
BANK
(229)
SHOKO
CHUKIN
BANK
(224)
SHOKO
CHUKIN
BANK
(206)
12
MEIJ
IYASUDA
LIF
E(2
96)
CHUO
MIT
-SUI
TRUST
&BANKIN
G(3
25)
CHUO
MIT
-SUI
TRUST
&BANKIN
G(3
34)
CHUO
MIT
-SUI
TRUST
&BANKIN
G(3
31)
SHOKO
CHUKIN
BANK
(274)
MEIJ
IYASUDA
LIF
E(1
56)
MEIJ
IYASUDA
LIF
E(1
68)
MEIJ
IYASUDA
LIF
E(1
80)
13
DBJ
(293)
MEIJ
IYASUDA
LIF
E(2
75)
SHOKO
CHUKIN
BANK
(302)
SHOKO
CHUKIN
BANK
(331)
MEIJ
IYASUDA
LIF
E(1
95)
MIZ
UHO
TRUST
&BANKIN
G(1
45)
YOKOHAM
ABANK
(157)
YOKOHAM
ABANK
(162)
14
YOKOHAM
ABANK
(258)
SHOKO
CHUKIN
BANK
(274)
MEIJ
IYASUDA
LIF
E(2
69)
MEIJ
IYASUDA
LIF
E(2
49)
YOKOHAM
ABANK
(191)
YOKOHAM
ABANK
(137)
MIZ
UHO
TRUST
&BANKIN
G(1
47)
MIZ
UHO
TRUST
&BANKIN
G(1
51)
15
DAIICHI
LIF
E(2
19)
YOKOHAM
ABANK
(258)
YOKOHAM
ABANK
(247)
YOKOHAM
ABANK
(247)
MIZ
UHO
TRUST
&BANKIN
G(1
74)
CHIB
ABANK
(99)
FUKUOKA
BANK
(106)
CHIB
ABANK
(108)
16
MIZ
UHO
TRUST
&BANKIN
G(2
02)
MIZ
UHO
TRUST
&BANKIN
G(1
99)
MIZ
UHO
TRUST
&BANKIN
G(1
85)
MIZ
UHO
TRUST
&BANKIN
G(1
87)
ORIX
(122)
FUKUOKA
BANK
(99)
CHIB
ABANK
(104)
FUKUOKA
BANK
(106)
17
SHOKO
CHUKIN
BANK
(202)
DAIICHI
LIF
E(1
73)
DAIICHI
LIF
E(1
43)
JOYO
BANK
(135)
JOYO
BANK
(122)
IYO
BANK
(97)
AOZORA
BANK
(96)
JOYO
BANK
(102)
18
AOZORA
BANK
(142)
ORIX
(131)
ORIX
(127)
CHIB
ABANK
(131)
CHIB
ABANK
(118)
AOZORA
BANK
(96)
IYO
BANK
(95)
SHIN
SEI
BANK
(94)
19
EXCEPT
OTHER
FI(1
37)
EXCEPT
OTHER
FI(1
29)
CHIB
ABANK
(125)
SHIG
ABANK
(130)
IYO
BANK
(110)
JOYO
BANK
(94)
JOYO
BANK
(93)
IYO
BANK
(93)
20
SUM
ITOM
OLIF
E(1
33)
CHIB
ABANK
(128)
SHIZ
UOKA
BANK
(123)
SHIZ
UOKA
BANK
(126)
UNKNOW
NPRI-
VATE
FI(1
10)
CREDIT
FED
OF
AGRICOOP
(86)
YAM
AGUCHI
BANK
(88)
HIR
OSHIM
ABANK
(89)
Note
s:Figuresin
parenth
eses
indicate
deg
reecentrality.BTMU
isBankofTokyo-M
itsu
bishiUFJ;SMBC
isSumitomoMitsu
iBanking
Corp
oration;MUTB
isMitsu
bishiUFJTru
st&
Banking;andDBJis
Dev
elopmen
tBankofJapan.
29
Figure 8: Degree time transition by institution
Notes: NOCHU is Norinchukin Bank and SHOCHU is Shoko Chukin Bank. The four panels show thedegree time transition for 16 selected institutions.
10 with Table 4, the transition matrix for FY2008–FY2009 stands out withregard to large downgrading probabilities, such as 43.30% (marked in gray)for a downgrade of a BB rating to default and 50.00% (marked in gray) fora downgrade of a B rating to default.
In addition, loss given default (LGD) is assumed to be 100% because,in terms of network structure, the usual lending relationships between adefaulted firm and its lending institutions are interrupted after default; thus,recovery takes approximately three to five years. The evaluation time pointand risk horizon are assumed to be the end of March 2019 and the endof March 2020 respectively. The other parameters (e.g., factor correlationmatrix) in Table 3 are assumed to be the same as those at the end of March2016.
Consequently, because many defaults occur, the lending network becomesmuch sparser. Table 11 indicates the number of defaulted firms, the numberof defaulted contracts (i.e., a firm’s default corresponds to a default for eachone of its banks), and VaR and ES by dependence structure on the value
30
SHINSEI BANK
AOZORA BANK
EDION
HULIC
NIPPON PAPER INDUSTRIES
MITSUBISHI MATERIALS
OKI ELECTRIC INDUSTRY
IWATANI
AEON FINANCIAL SERVICE
TOKYO DOME
MIZUHO BANK
BTMU
SMBC
RESONA BANK
MIZUHO CORP BANK
SAITAMA RESONA BANK
77 BANK
GUNMA BNAK
ASHIKAGA BANK
JOYO BANK
CHIBA BANK
TOMIN BANK
YOKOHAMA BANK
DAISHI BANK
82 BANK
HOKURIKU BANK
HOKKOKU BANK
SHIZUOKA BANK
OGAKI KYORITSU BANK 16 BANK
MIE BANK
105 BANK
SHIGA BANK
KYOTO BANK
IKEDA SENSHU BANK
NANTO BANK
KIYO BANK
SANIN GODO BANK
CHUGOKU BANK
HIROSHIMA BANK
YAMAGUCHI BANK
114 BANK
IYO BANK
FUKUOKA BANK
HIGO BANK
NISHINIHON CITY BANK
MUTB MIZUHO TRUST& BANKING
CHUO MITSUI TRUST& BANKING
SUMITOMO MITSUI TRUST BANK
NIPPON LIFE
DAIICHI LIFE
FUKOKU LIFE
ASAHI LIFE
MEIJI YASUDA LIFEMITSUI LIFE
SUMITOMO LIFE
AICHI BANK
NAGOYA BANK
CHUKYO BANK
KANSAI URBAN BANK
MINATO BANK
SHINKIN CENTRAL BANK
SHINKIN BANK
NORINCHUKIN BANK
CREDIT FED OF AGRI COOP
SHOKO CHUKIN BANK
UNKNOWN PRIVATE FI
DBJ
OTHER PUBLIC FI
OTHER GOVERNMENT FI
EXCEPT OTHER FI
UNKNOWN
Figure 9: Directed graph of degrees over 40, end of March 2009 (just afterthe bankruptcy of Lehman Brothers)
Note: These graphs are drawn in accordance with the Fruchterman–Reingold algorithm.
31
SHINSEI BANK
AOZORA BANK
HULIC
SAMTY
OKI ELECTRIC INDUSTRY
FINANCIAL PRODUCTS GROUP
HITACHI CAPITAL
AEON MALL
KYUSHU ELECTRIC POWER
MIZUHO BANK
BTMU
SMBC
RESONA BANK
77 BANK
GUNMA BNAK
ASHIKAGA BANK
JOYO BANK
CHIBA BANK
TOMIN BANK
YOKOHAMA BANK
DAISHI BANK
82 BANK
HOKURIKU BANK
SHIZUOKA BANK
OGAKI KYORITSU BANK
16 BANK
MIE BANK
105 BANK
SHIGA BANK
KYOTO BANK
IKEDA SENSHU BANK
NANTO BANK
SANIN GODO BANK
CHUGOKU BANK
HIROSHIMA BANK
YAMAGUCHI BANK
114 BANK
IYO BANK
FUKUOKA BANK
HIGO BANK
NISHINIHON CITY BANK
MUTB
MIZUHO TRUST& BANKING
SUMITOMO MITSUI TRUST BANK
NIPPON LIFE
DAIICHI LIFE
MEIJI YASUDA LIFE
HIGASHI-NIPPON BANK
AICHI BANK
NAGOYA BANK
MINATO BANK
SHINKIN CENTRAL BANK
SHINKIN BANK
NORINCHUKIN BANK
CREDIT FED OF AGRI COOP
SHOKO CHUKIN BANK
UNKNOWN PRIVATE FI
DBJ
OTHER PUBLIC FI
EXCEPT OTHER FI
UNKNOWN
Figure 10: Directed graph of degrees over 40, end of March 2016
Note: These graphs are drawn in accordance with the Fruchterman–Reingold algorithm.
32
SHINSEI BANK
AOZORA BANK
HULIC
NIPPON PAPER INDUSTRIES
MITSUBISHI MATERIALS
HITACHI CONST MACHINERY
OKI ELECTRIC INDUSTRY
AEON FINANCIAL SERVICE
ORIX
TOKYO DOME
MIZUHO BANK
BTMU
SMBC
RESONA BANK
MIZUHO CORP BANK
SAITAMA RESONA BANK
77 BANK
GUNMA BNAK
ASHIKAGA BANK
JOYO BANK
CHIBA BANK
TOMIN BANK
YOKOHAMA BANK
DAISHI BANK
82 BANK
HOKURIKU BANK
HOKKOKU BANK
SHIZUOKA BANK
OGAKI KYORITSU BANK
16 BANK
MIE BANK
105 BANK
SHIGA BANK
KYOTO BANK
IKEDA SENSHU BANK
NANTO BANK
KIYO BANK
SANIN GODO BANK
CHUGOKU BANK
HIROSHIMA BANK
YAMAGUCHI BANK
114 BANK
IYO BANK
FUKUOKA BANK
HIGO BANK
NISHINIHON CITY BANK
MUTB
MIZUHO TRUST& BANKING
CHUO MITSUI TRUST& BANKING
SUMITOMO MITSUI TRUST BANK
NIPPON LIFE
DAIICHI LIFE
ASAHI LIFE
MEIJI YASUDA LIFE
MITSUI LIFE
SUMITOMO LIFE
HOKUYO BANK
AICHI BANK
NAGOYA BANK
CHUKYO BANK
KANSAI URBAN BANK
MINATO BANK
SHINKIN CENTRAL BANK
SHINKIN BANK
NORINCHUKIN BANK
CREDIT FED OF AGRI COOP
SHOKO CHUKIN BANK
UNKNOWN PRIVATE FI
DBJ
OTHER PUBLIC FI
OTHER GOVERNMENT FI
EXCEPT OTHER FI
UNKNOWN
SHINSEI BANK
AOZORA BANK
HULIC
MITSUBISHI MATERIALS
OKI ELECTRIC INDUSTRY
AEON FINANCIAL SERVICE
ORIX
MIZUHO BANK
BTMU
SMBCRESONA BANK
MIZUHO CORP BANK
SAITAMA RESONA BANK
77 BANK
GUNMA BNAK
ASHIKAGA BANK
JOYO BANK
CHIBA BANK
TOMIN BANK
YOKOHAMA BANK
DAISHI BANK
82 BANK
HOKURIKU BANK
HOKKOKU BANK
SHIZUOKA BANK
OGAKI KYORITSU BANK
16 BANK
105 BANK
SHIGA BANK
KYOTO BANK
IKEDA SENSHU BANK
NANTO BANK
KIYO BANK
SANIN GODO BANK
CHUGOKU BANK
HIROSHIMA BANK
YAMAGUCHI BANK
114 BANK
IYO BANK
FUKUOKA BANK
HIGO BANK
NISHINIHON CITY BANK
MUTB
MIZUHO TRUST& BANKING
CHUO MITSUI TRUST& BANKING
SUMITOMO MITSUI TRUST BANK
NIPPON LIFE
DAIICHI LIFE
MEIJI YASUDA LIFE
MITSUI LIFE
SUMITOMO LIFE
AICHI BANK
NAGOYA BANK
CHUKYO BANK
KANSAI URBAN BANK
MINATO BANK
SHINKIN CENTRAL BANK
SHINKIN BANK
NORINCHUKIN BANK
CREDIT FED OF AGRI COOP
SHOKO CHUKIN BANK
UNKNOWN PRIVATE FI
DBJ
OTHER PUBLIC FI
OTHER GOVERNMENT FI
EXCEPT OTHER FI
UNKNOWN
SHINSEI BANK
AOZORA BANK
HULIC
MITSUBISHI MATERIALS
OKI ELECTRIC INDUSTRY
AEON FINANCIAL SERVICE
ORIXKYUSHU ELECTRIC POWER
MIZUHO BANK
BTMU
SMBC
RESONA BANK
MIZUHO CORP BANK
SAITAMA RESONA BANK
77 BANK
GUNMA BNAK
ASHIKAGA BANK
JOYO BANK
CHIBA BANK
TOMIN BANK
YOKOHAMA BANK
DAISHI BANK
82 BANK
HOKURIKU BANK
HOKKOKU BANK
SHIZUOKA BANK
OGAKI KYORITSU BANK
16 BANK
MIE BANK
105 BANK
SHIGA BANK
KYOTO BANK
IKEDA SENSHU BANK
NANTO BANK
KIYO BANK
SANIN GODO BANK
CHUGOKU BANK
HIROSHIMA BANK
YAMAGUCHI BANK
114 BANK
IYO BANK
FUKUOKA BANK
HIGO BANK
NISHINIHON CITY BANK
MUTB
MIZUHO TRUST& BANKING
CHUO MITSUI TRUST& BANKING
SUMITOMO MITSUI TRUST BANK
ORIX BANK
NIPPON LIFE
DAIICHI LIFE
MEIJI YASUDA LIFE
MITSUI LIFE
SUMITOMO LIFE
HIGASHI-NIPPON BANK
AICHI BANK
NAGOYA BANK
CHUKYO BANK
KANSAI URBAN BANK
MINATO BANK
SHINKIN CENTRAL BANK
SHINKIN BANK
NORINCHUKIN BANK
CREDIT FED OF AGRI COOP
SHOKO CHUKIN BANK
OTHER PRIVATE FI
UNKNOWN PRIVATE FI
DBJ
OTHER PUBLIC FI
OTHER GOVERNMENT FI
EXCEPT OTHER FI
UNKNOWN
SHINSEI BANK
AOZORA BANK
HULIC
MITSUBISHI MATERIALS
AEON FINANCIAL SERVICE
ORIX
KYUSHU ELECTRIC POWER
MIZUHO BANK
BTMU
SMBC
RESONA BANK
MIZUHO CORP BANK
77 BANK
GUNMA BNAK
ASHIKAGA BANK
JOYO BANK
CHIBA BANK
TOMIN BANK
YOKOHAMA BANK
DAISHI BANK
82 BANK
HOKURIKU BANK
HOKKOKU BANK
SHIZUOKA BANK
OGAKI KYORITSU BANK
16 BANK
105 BANK
SHIGA BANK
KYOTO BANK
IKEDA SENSHU BANK
NANTO BANK
KIYO BANK
SANIN GODO BANK
CHUGOKU BANK
HIROSHIMA BANK
YAMAGUCHI BANK
114 BANK
IYO BANK
FUKUOKA BANK
NISHINIHON CITY BANK
MUTB
MIZUHO TRUST& BANKING
SUMITOMO MITSUI TRUST BANK
ORIX BANK
NIPPON LIFE
DAIICHI LIFE
MEIJI YASUDA LIFE
MITSUI LIFE
SUMITOMO LIFE
HIGASHI-NIPPON BANK
AICHI BANK
NAGOYA BANK
KANSAI URBAN BANK
MINATO BANK
SHINKIN CENTRAL BANK
SHINKIN BANK
NORINCHUKIN BANK
CREDIT FED OF AGRI COOP
SHOKO CHUKIN BANK
OTHER PRIVATE FI
UNKNOWN PRIVATE FI
DBJ
OTHER PUBLIC FI
EXCEPT OTHER FI
UNKNOWN
SHINSEI BANK
AOZORA BANK
HULIC
KYUSHU ELECTRIC POWER
MIZUHO BANK
BTMU
SMBC
RESONA BANK
77 BANK
GUNMA BNAK
JOYO BANK
CHIBA BANK
TOMIN BANK
YOKOHAMA BANK
82 BANK
HOKURIKU BANK
SHIZUOKA BANK
OGAKI KYORITSU BANK
16 BANK
105 BANK
SHIGA BANK
KYOTO BANK
IKEDA SENSHU BANK
NANTO BANK
KIYO BANK
SANIN GODO BANK
CHUGOKU BANK
HIROSHIMA BANK
YAMAGUCHI BANK
114 BANK
IYO BANK
FUKUOKA BANK
NISHINIHON CITY BANK
MUTB
MIZUHO TRUST& BANKING
SUMITOMO MITSUI TRUST BANK
NIPPON LIFE
DAIICHI LIFE
MEIJI YASUDA LIFE
HIGASHI-NIPPON BANK
AICHI BANK
NAGOYA BANK
KANSAI URBAN BANK
MINATO BANK
SHINKIN CENTRAL BANK
SHINKIN BANK
NORINCHUKIN BANK
CREDIT FED OF AGRI COOP
SHOKO CHUKIN BANK
OTHER PRIVATE FI
UNKNOWN PRIVATE FI
DBJ
OTHER PUBLIC FI
EXCEPT OTHER FI
UNKNOWN
SHINSEI BANK
AOZORA BANK
HULIC
OKI ELECTRIC INDUSTRY
FINANCIAL PRODUCTS GROUP
HITACHI CAPITAL
AEON MALL
KYUSHU ELECTRIC POWER
MIZUHO BANK
BTMU
SMBC
RESONA BANK
77 BANK
GUNMA BNAK
ASHIKAGA BANK
JOYO BANK
CHIBA BANK
TOMIN BANK
YOKOHAMA BANK
DAISHI BANK
82 BANK
HOKURIKU BANK
SHIZUOKA BANK
OGAKI KYORITSU BANK
16 BANK
MIE BANK
105 BANK
SHIGA BANK
KYOTO BANK
IKEDA SENSHU BANK
NANTO BANK
KIYO BANK
SANIN GODO BANK
CHUGOKU BANKHIROSHIMA BANK
YAMAGUCHI BANK
114 BANK
IYO BANK
FUKUOKA BANK
NISHINIHON CITY BANK
MUTB
MIZUHO TRUST& BANKING
SUMITOMO MITSUI TRUST BANK
NIPPON LIFE
DAIICHI LIFE
MEIJI YASUDA LIFE
HIGASHI-NIPPON BANK
AICHI BANK
NAGOYA BANK
KANSAI URBAN BANK
MINATO BANK
SHINKIN CENTRAL BANK
SHINKIN BANK
NORINCHUKIN BANK
CREDIT FED OF AGRI COOP
SHOKO CHUKIN BANK
OTHER PRIVATE FI
UNKNOWN PRIVATE FI
DBJ
OTHER PUBLIC FI
EXCEPT OTHER FI
UNKNOWN
Figure 11: Directed graphs of degrees over 40
Notes: The six panels show directed graphs of firm nodes over 40 degrees at the end of March 2010 andMarch 2011 from the upper-left panel to the upper-right panel; at the end of March 2012 andMarch 2013 from the middle-left panel to the middle-right panel; and at the end of March 2014and March 2015 from the lower-left panel to the lower-right panel.
33
Table 10: Transition matrix with averaged R&I’s annual rating migrationrates for FY2008–FY2009
AAA AA A BBB BB B CCC-C DefaultAAA 0.00 100.00 0.00 0.00 0.00 0.00 0.00 0.00AA 0.00 79.50 18.35 1.10 0.70 0.35 0.00 0.00A 0.00 3.90 88.00 7.60 0.00 0.00 0.00 0.50BBB 0.00 0.00 14.85 78.10 4.30 0.00 0.00 2.75
BB 0.00 0.00 20.00 13.35 23.35 0.00 0.00 43.30
B 0.00 0.00 0.00 0.00 50.00 0.00 0.00 50.00CCC-C 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00Default 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00
Notes: Stressed is loaded to the credit rating migration matrix for one year from a futuretime point to a risk horizon. The matrix is provided as a one-year average for the2008 and 2009 cohorts provided by R&I.
distribution of lending assets: Gaussian copula or t5 copula. As a result,defaulted firms account for 1.7% of all Japanese listed firms in terms ofGaussian copula dependence and 2.2% in terms of t5 copula dependence.Stressed VaR and ES are over around 500 times greater than normal. Inparticular, the t5 copula ES of JPY 38 trillion is an enormous sum in theJapanese financial system.
At the same time, Figure 12 indicates directed graphs of degrees over 40,pertaining to a scenario of a 99.9% confidence level of portfolio values at arisk horizon of all non-defaulted firms. The left panel of Figure 12 assumesGaussian copula dependence and the right panel assumes t5 copula depen-dence. Comparing both panels with Figure 11, each panel shows a sparsernetwork structure. In particular, because t5 copula brings tail dependencyinto the value distribution, t5 VaR and t5 ES are much larger than Gaussiancopula VaR and Gaussian ES respectively.
6. Conclusions
This study contributes to the literature by analyzing credit rating migra-tion risk in Japan’s corporate lending market.
First, in corporate credit risk management, the study evaluated the creditrisk exposure for all Japanese listed firms. Following the credit migrationapproach, a firm’s credit risk exposure changes depending on its corporate
34
Table 11: Results calculated using the transition matrix for FY2008–FY2009(in JPY billion)
Defaulted firms Defaulted contracts VaR ESNumber % Number % Amount % Amount %
GC 38 1.7 201 1.5 17,181 530 20,748 547t5-C 48 2.2 374 2.9 31,254 579 37,650 501
Notes: GC is Gaussian copula and t5-C is t copula with degree of freedom five. Becausea firm generally borrows money from institutions, one firm’s default results inlosses for such institutions. Each percentage in the columns for “Defaulted firms”and “Defaulted contracts” denotes a multiple of the number of each total number.Each percentage in the columns for “VaR” and “ES” denotes a multiple of therisk amounts in normal times.
rating. The values of the outstanding lending of mega banks also substan-tially reduced just after the bankruptcy of Lehman Brothers. By contrast,the outstanding values for life insurers increased after FY2009. The analyti-cal results show that banks are affected by the capital requirement of Basel IIand III, whereas life insurers aimed to improve their investment performanceduring the studied period.
Second, this study measured the lending portfolio credit risk for majorbanks and other large banks, and major life insurers. The risk measures usedare VaR and ES. In particular, ES is expected to ensure the prudent captureof tail risk and has actually been introduced in insurer solvency regulation,as illustrated by the Swiss Solvency Test. In addition, the choice of copulais critical for correctly measuring the dependence between systematic riskfactors.
Third, this study ranked the industry sectors in accordance with creditrisk exposure and lending portfolio credit risk. In terms of credit risk ex-posure, sectors such as Wholesale Trade, Other Financing Business, Land& Transportation, Electric Power & Gas, and Real Estate were ranked inthe top ten. By contrast, sectors such as Electric Power & Gas, Chemicals,and Information & Communication, which are all ranked in the top ten interms of credit risk exposure, are outside the ranking for all risk measures.Although these sectors have large credit risk exposure, credit risk amountsmeasured by VaR and ES are all small because of the small volatilities ofportfolio value distributions.
35
SHINSEI BANK
AOZORA BANK
HULIC
SAMTY
OKI ELECTRIC INDUSTRY
FINANCIAL PRODUCTS GROUP
HITACHI CAPITAL
AEON MALL
KYUSHU ELECTRIC POWER
MIZUHO BANK
BTMU
SMBC
RESONA BANK
77 BANK
GUNMA BNAK
ASHIKAGA BANK
JOYO BANK
CHIBA BANK
TOMIN BANK
YOKOHAMA BANK
DAISHI BANK
82 BANK
HOKURIKU BANK
SHIZUOKA BANK
OGAKI KYORITSU BANK
16 BANK
MIE BANK
105 BANK
SHIGA BANK
KYOTO BANK
IKEDA SENSHU BANK
NANTO BANK
SANIN GODO BANK
CHUGOKU BANK
HIROSHIMA BANK
YAMAGUCHI BANK
114 BANK
IYO BANK
FUKUOKA BANK
HIGO BANK
NISHINIHON CITY BANK
MUTB
MIZUHO TRUST& BANKING
SUMITOMO MITSUI TRUST BANK
NIPPON LIFE
DAIICHI LIFE
MEIJI YASUDA LIFE
HIGASHI-NIPPON BANK
AICHI BANK
NAGOYA BANK
MINATO BANK
SHINKIN CENTRAL BANK
SHINKIN BANK
NORINCHUKIN BANK
CREDIT FED OF AGRI COOP
SHOKO CHUKIN BANK
UNKNOWN PRIVATE FI
DBJ
OTHER PUBLIC FI
EXCEPT OTHER FI
UNKNOWN
SHINSEI BANK
AOZORA BANK
HULIC
SAMTY
OKI ELECTRIC INDUSTRY
FINANCIAL PRODUCTS GROUP
HITACHI CAPITAL
AEON MALL
KYUSHU ELECTRIC POWER
MIZUHO BANK
BTMU
SMBC
RESONA BANK
77 BANK
GUNMA BNAK
ASHIKAGA BANK
JOYO BANK
CHIBA BANK
TOMIN BANK
YOKOHAMA BANK
DAISHI BANK
82 BANK
HOKURIKU BANK
SHIZUOKA BANK
OGAKI KYORITSU BANK
16 BANK
MIE BANK
105 BANK
SHIGA BANK
KYOTO BANK
IKEDA SENSHU BANK
NANTO BANK
SANIN GODO BANK
CHUGOKU BANK
HIROSHIMA BANK
YAMAGUCHI BANK
114 BANK
IYO BANK
FUKUOKA BANK
HIGO BANK
NISHINIHON CITY BANK
MUTB
MIZUHO TRUST& BANKING
SUMITOMO MITSUI TRUST BANK
NIPPON LIFE
DAIICHI LIFE
MEIJI YASUDA LIFE
HIGASHI-NIPPON BANK
AICHI BANK
NAGOYA BANK
MINATO BANK
SHINKIN CENTRAL BANK
SHINKIN BANK
NORINCHUKIN BANK
CREDIT FED OF AGRI COOP
SHOKO CHUKIN BANK
UNKNOWN PRIVATE FI
DBJ
OTHER PUBLIC FI
EXCEPT OTHER FI
UNKNOWN
Figure 12: Directed graphs of degrees over 40, pertaining to a scenario ofa 99.9% confidence level of portfolio values at a risk horizon (left panel:Gaussian copula; right panel: t5 copula)
Note: The two panels show the stressed Gaussian and t5 copula distributions of portfolio values for allinstitutions at the end of March 2019.
Fourth, this study analyzed the network structure of corporate lendingamong bank-to-listed firms and insurer-to-listed firms in Japan’s lending mar-ket using major centrality measures. Banks and insurers play a central rolein terms of degree centrality. However, degree centrality decreased graduallyafter the global financial crisis. This may mean a decrease in the number ofcounterparties.
Fifth, this study conducted a stress test in terms of network structure.Because 1.7% of all Japanese listed firms defaulted in terms of Gaussiancopula dependence and 2.2% in terms of t5 copula dependence, the networkstructure became much sparser.
Finally, this study’s analyses on credit rating migration risk and inter-connectedness in a lending network can serve as warnings to related entitiessuch as financial institutions, supervisory authorities, and firms about riskperception.
To conclude, because our data are restricted to the Japanese market, itwould be effective to apply our methodology to other financial markets for
36
further studies.
Acknowledgments
This study was supported by the Japan Society for the Promotion ofScience (JSPS) KAKENHI [Grants-in-Aid for Scientific Research, 17K03813].Such assistance is sincerely appreciated.
References
[1] Abbassi, P., Brownlees, C., Hans, C., Podlich, N., 2017. Credit riskinterconnectedness: What does the market really know? Journal ofFinancial Stability. 29, 1-12.
[2] Alves, I., Ferrari, S., Franchini, P., Heam, J.-C., Jurca, P., Langfield,S., Laviola, S., Liedorp, F., Sanchez, A., Tavolaro, S., Vuillemey, G.,2013. The structure and resilience of the European interbank market.Occasional Paper Series, European Systemic Risk Board. 3(September).
[3] Basel Committee on Banking Supervision (BCBS), 2005. Internationalconvergence of capital measurement and capital standards: A revisedframework.
[4] Basel Committee on Banking Supervision (BCBS), 2016. StandardTLAC holdings amendments to the Basel III standard on the defini-tion of capital.
[5] Boss, M., Elsinger, H., Summer, M., Thurner, S., 2004. An empiricalanalysis of the network structure of the Austrian interbank market. Fi-nancial Stability Report, Oesterreichische Nationalbank. 77–87.
[6] Breuer, T., Jandacka, M., Mencıa, J., Summer, M., 2012. A systematicapproach to multi-period stress testing of portfolio credit risk. Journalof Banking & Finance. 36, 332–340.
[7] Cherubini, U., Mulinacci, S., Gobbi, F., Romagnoli, S., 2011. DynamicCopula Methods in Finance. West Sussex: John Wiley & Sons.
[8] Chuluun, T., Prevost, A., Upadhyay, A., 2017. Firm network structureand innovation. Journal of Corporate Finance. 44, 193–214.
37
[9] Cocco, J., Gomes, F., Martins, N., 2009. Lending relationships in theinterbank market. Journal of Financial Intermediation. 18(1), 24–48.
[10] Crouhy, M., Galai, D., Mark, R., 2000. A comparative analysis of currentcredit risk models. Journal of Banking and Finance. 24, 59–117.
[11] De Masi, G., Gallegati, M., 2012. Bank–firms topology in Italy. Empir-ical Economics. 43, 851–866.
[12] Eisenberg, L., Noe, T., 2001. Systemic risk in financial systems. Man-agement Science. 47, 236–49.
[13] Elsinger, H., Lehar, A., Summer, M., 2006. Using market information forbanking systems. International Journal of Central Banking. 27, 137–165.
[14] Garmaise, M.J., Moskowitz, T.J., 2003. Informal financial networks:Theory and Evidence. Review of Financial Studies. 16(4), 1007–1040.
[15] Godlewski, C. J., Sanditov, B., Burger-Helmchen, T., 2012. Bank lend-ing networks, experience, reputation, and borrowing costs: Empiricalevidence from the French syndicated lending market. Journal of Busi-ness Finance & Accounting. 39(1) & (2), 113–140.
[16] Gordy, M., 2000. A comparative anatomy of credit risk models. Journalof Banking and Finance. 24, 119–149.
[17] Gupton, G., Finger, C., Bhatia, M., 1997. CreditMetrics – TechnicalDocument. J. P. Morgan, New York.
[18] Ha laj, G., Kochanska, U., Kok, C., 2015. Emergence of the EU corporatelending network. Journal of Network Theory in Finance. 1(1), 33–76.
[19] Haldane, A.G., May, R.M., 2011. Systemic risk in banking ecosystems.Nature. 469, 351–355.
[20] Henry, J., Kok, C., 2013. A macro stress testing framework for assessingsystemic risks in the banking sector. Occasional Paper Series, EuropeanCentral Bank. 152 (October).
[21] Jackson, M.O., 2010. Social and Economic Networks. New Jersey:Princeton University Press.
38
[22] Jarrow, R.A., Lando, D., Turnbull, S., 1997. A Markov chain model forthe term structure of credit risk spreads. Review of Financial Studies.10(2), 481–523.
[23] Kanno, M., 2015a. Assessing systemic risk using interbank exposures inthe global banking system. Journal of Financial Stability. 20, 105–130.
[24] Kanno, M., 2015b. Macro stress test for credit risk. Journal of RiskFinance. 16(5), 554–574.
[25] Kanno, M., 2015c. The network structure and systemic risk in theJapanese interbank market. Japan and the World Economy. 36, 102–112.
[26] Kanno, M., 2018a. Bank-Insurer-Firm tripartite interconnectedness ofcredit risk exposures in a cross-shareholding network. Risk Management.20(4), 273–303.
[27] Kanno, M., 2018b. Network structures and credit risk in the cross-shareholdings among listed Japanese companies. Japan and the WorldEconomy. https://doi.org/10.1016/j.japwor.2018.09.003.
[28] Krause, A., Giansante, S., 2012. Interbank lending and the spread ofbank failures: A network model of systemic risk. Journal of EconomicBehavior & Organization. 83(3), 583–608.
[29] Li, H., An, H., Gao, X., Huang, J., Xu, Q., 2014. On the topolog-ical properties of the cross-shareholding networks of listed companiesin China: Taking shareholders’ cross-shareholding relationships into ac-count. Physica A. 406, 80–88.
[30] Lux, T., 2016. A model of the topology of the bank–firm credit networkand its role as channel of contagion. Journal of Economic Dynamics andControl. 66, 36–53.
[31] Newman, M.E.J., 2010. Networks. Oxford: Oxford University Press.
[32] Paltalidis, N., Gounopoulos, D., Kizys, R., Koutelidakis, Y., 2015.Transmission channels of systemic risk and contagion in the Europeanfinancial network. Journal of Banking & Finance. 61, S36–S52.
39
[33] R&I, 2016. Japanese corporate default rates and rating transition ma-trices (FY1978–FY2015).
[34] Schuermann, T., 2014. Stress testing banks. International Journal ofForecasting. 30, 717–728.
[35] Silva, T.C., Alexandre, M., Tabak, B.M., 2018. Bank lending and sys-temic risk: A financial-real sector network approach with feedback. Jour-nal of Financial Stability. 38, 98–118.
[36] Tsaig, Y., Levy, A., Wang, Y., 2011. Analyzing the impact of creditmigration in a portfolio setting. Journal of Banking & Finance. 35, 3145–3157.
[37] Varotto, S., 2012. Stress testing credit risk: The Great Depression sce-nario. Journal of Banking & Finance. 36, 3133–3149.
[38] Wang, L., Wang, S., 2012. Endogenous networks in investment syndica-tion. Journal of Corporate Finance. 18, 640–663.
[39] Wu, W.-S., Chang, H.-H., Suardi, S., Chang, Y., 2013. The cascadeeffect on lending conditions: Evidence from the syndicated loan market.Journal of Business Finance & Accounting. 40(9) & (10), 1247–1275.
40